Study of Filter-based Neuromorphic Photonic Reservoir Computing for Signal Equalization in 224Gbps Sub-carrier Modulation IM-DD Short Reach Optical Fiber Communication System

Abstract

The ever-increasing requirements for bandwidth in edge places higher demands on the transmission capacity and data rate of short-reach intensity-modulation and direct-detection (IM/DD) optical fiber communication systems. Advanced digital signal processing (DSP), such as neural network (NN), is verified to be a good way to improve system performance, but the complicated DSP process always means high power consumption and slow processing speed. Reservoir Computing (RC) is a machine learning algorithm suitable for time-series-based problem, which has a faster computing speed than recurrent NN (RNN). The inherent randomness of RC makes us find its potential of signal equalization in all-optical domain. In this paper, we numerically studied a neuromorphic photonic RC signal processing scheme in IM/DD system with low hardware complexity, and realize the all-optical RC through two sets of optical filter nodes. Subcarrier modulation (SCM) signal is applied to study the filter-based neuromorphic photonic RC scheme, in comparison to traditional equalization methods. Simulation results show that the photonic RC equalization can bring orders of magnitude improvement in BER over traditional schemes, and the performances of different Quadrature Amplitude Modulation (QAM) formats are also studied. Finally, the architecture implementation of photonics RC for 224Gbps SCM signal over 80km standard single-mode fiber (SSMF) transmission in C-band is numerically demonstrated.